Computational studies of consciousness 2009 4 2 Yonsei

Computational studies of consciousness 2009. 4. 2. Yonsei University Dept. of Computer Science Young-Seol, Lee 0

Outline • Introduction and overview • Some history and those involved • The aims of computational models of consciousness • The challenge of phenomenology: is there a hard problem? • Depiction: the key mechanism for conscious representation • Current Perspectives on five axioms • The mechanistic use of the axioms: the kernel architecture • The kernel architecture and visual awareness • Unstable vision 1

Outline • Is an organism conscious? • Animal consciousness • Higher order thought • The kernel architecture, experience, sleep and dreaming • Application to machines • Summary and conclusions 2

Introduction and Overview • Professional philosopher’s perspective – The link between computational material and consciousness • Difficult and puzzling topic – Making machines that are conscious like ourselves • Particularly discourage – Computational modeling • The part of the philosophical process of understanding • Computational studies – Extracting the physical features that support a real material • Ex) a real hurricane in a real world – Creating it in a virtual world (in computer) • Ex) a virtual hurricane in a virtual world – The purpose of virtual hurricane in a virtual world • How a real hurricane will behave in a real world 3

Introduction and Overview • Hurricane → consciousness – Extracting physical features that support real consciousness • Real consciousness in a real brain – Creating it in a virtual world (in computer) • Creating virtual consciousness in a virtual world – Physical features of consciousness → difficult • Aim of the paper – Designing a machine that will be ‘conscious of a real world in the way we are’is not aim – Discussion of the very first step of extracting important mechanisms necessary for the building of a virtual machine – Breaking consciousness down • Five distinct ‘axioms’to facilitate the design process – Looking at the current nature of the paradigm of ‘Macine Consciousness’ 4

Introduction and Overview • Discussion of some important puzzles related to the mechanisms of consciousness – Why does vision play tricks sometimes? – How does one check for the presence of consciousness? – Are animals conscious? – Is there higher order consciousness? – What happens in unconscious moments? 5

Some history and those involved • International conference on Artificial Neural Networks (1992) – The discovery of the neural correlates of consciousness • Suggestion of future challenge for neural network researchers • A small workshop sponsored by Swartz foundation (2001, May) – The paradigm of machine consciousness • Establishment among international workers • “one day computers or robots could be conscious. … we know of no fundamental law or principle operating in this universe that forbids the existence of subjective feelings in artifacts designed or evolved by humans” • Franklin (2003) – Intelligent distribution agent based on Baars’Global workspace • Creating a system the users of which believe that they are dealing with an entity that is aware of their needs 6

Some history and those involved • Franklin (2003) – Intelligent distribution agent based on Baars’Global workspace • A ‘functional’approach to models which create systems that appear to be conscious through their behavior • Aleksander and Dunmall (2003) – ‘phenomenological’ approaches • Concerning mechanisms that may be needed to generate internal sensations • Major issues at conferences since 2003 – Association for the Scientific Study of Consciousness (ASSC) – Artificial Intelligence and Simulation of Behavior (AISB) • The design of conscious machines (2006) • Machines with imagination (2005) • The generation of a synthetic phenomenology (2006) 7

The aims of computational models of consciousness • Extraction of the essential physical features of consciousness – Hypothesising such physical features using an axiomatic decomposition – Desired virtual organism conscious of a virtual world – Assessment of what this tells us about a real organism that is conscious of a real world • Motivation for machine models of consciousness – To understand in a constructive way – To be able to discuss formally some puzzling features of consciousness • unstable vision, illusions, test for being consciousness, unconsciousness … – To encourage those who work with conscious organisms to face the complexity of the brain in a formal way – To ask if implemented systems have new uses 8

The aims of computational models of consciousness • Consciousness (in this paper) – A definite product of the brain with five axiomatic features • Presence, imagination, attention, planning and emotions • Technological basis of this work – Neural architectures studied with a computationally efficient model of the neuron – Kernel architecture (KA) • Ensemble of five axiomatic features • Central mechanism that support the axioms → Depiction – Sensory pathways in the brain – Signals from the musculature of the body 9

The challenge of phenomenology: is there a hard problem? • Phenomenology – Personal, internal feelings of being conscious – Introspection will be the starting point for computational studies • Implication for computational studies – A link between experienced conscious states and observable states of some underlying physical computational mechanism – Computation (at this paper) • State development of an architecture that is controlled • The prototype model for such an architecture – Living brain 10

The challenge of phenomenology: is there a hard problem? • Implication for computational studies – Assertion 1 • To include phenomenology in a computational model of consciousness it is necessary to abstract the physical/informational properties of the brain that are hypothesised to determine conscious states and study these as structures that are virtual on a conventional computer architecture in order to check the validity of the hypotheses • Virtual architecture which is brain-like in broad essence – Chalmers (1996) • While work on the physical may be very interesting, it does not lead to an understanding of the phenomenological – Consciousness is free from physical process – To understand phenomenology • Start with an organism that is phenomenologically conscious 11

Depiction: the key mechanism for conscious representation • Depiction – A direct representation of where elements of the world are in the world which is encoded by the efforts of the mechanism to attend to such elements • Assertion 2 – The parts of a mechanism that sustain conscious experience can only do so if they are the product of a depictive process 12

Current perspective on five axioms • Dividing conscious experience into five axioms – Following introspective route – Aleksander and Dunmall (2003) • Presence – I feel that I am an entity in world that is outside of me • The feeling of being an empowered, active agent in a real sensory world • Imagination – I can recall previous sensory experience as a more or less degraded version of that experience. Driven by language I can imagine experiences I never had • Creating scenarios that might have been experienced or even ones that are not close to reality 13

Current perspective on five axioms • Attention – I am selectively conscious of the world outside of me and can select sensory events I wish to imagine • Sometimes what we choose to experience is automatic • Ex) bright flash, sound bang, etc. • Ex) identifying the make of a car – Bonnet, wheels, rear for seeing an emblem • Volition (previously called Planning) – I can imagine the results of taking actions and select an action I wish to take • Imagine taking the actions in succession • Prediction of the ensuing results even if they may not be very clear • Ex) choosing between going to a restaurant we know and going to a previously untried restaurant 14

Current perspective on five axioms • Emotion – I evaluate events and the expected results of actions according to criteria usually called emotions – Decision about which restaurant will be chosen → evaluation – Ex) eating stake in a steak house • Pleasure of the experience • Negative feeling about the intake of cholesterol – Ex) unknown restaurant • A fear of the unknown • Excitement of a new adventure 15

The mechanistic use of the axioms: the kernel architecture • Next question – How they(fix axioms) can interlock to provide a sensation of a unified consciousness ? • Kernel Architecture (KA) – An abstract physical structure composed of neural areas – The word‘kernel’ • It appears as part of many models that have been implemented in the past as is seen below 16

The mechanistic use of the axioms: the kernel architecture • Presence – Input of the neuron • Sensory signal • Some frame of reference in the world – Output of the neuron • Representative of where an element of the sensory world is located – Ex) binding problem • A small moving green triangle stimulates representational neurons in different parts of the visual cortex – A color area, a motion area and a shape area – How this integrates to provide a coherent sensation ? • The three separate neurons (each related to 3 kinds) – The sensation is one of overlap – Perceptual module in KA • Support mechanisms for Axiom 1 17

The mechanistic use of the axioms: the kernel architecture • Imagination – An example of the simplest imaginational act • Looking at an object in each side of the room – Left (object A), Right (object B) – Input of the neuron • The depiction in the perceptual module – Iconic learning • Transfer of input pattern Ai to the state variables of the memory module to become Ai’ • Ai and Ai’ need not be exactly same – Memory is less vivid than perception 18

The mechanistic use of the axioms: the kernel architecture • Attention – The perceptual automaton – Top state showing both A and B is a hazy depiction 19

The mechanistic use of the axioms: the kernel architecture • Volition and Emotion – Simple scenario (restaurant) • • • P 1 : depiction of the menu A 1 : a memory of the first item (pizza) R 1 : experience of eating the pizza E 1 j, e 1 k : Emotions accompanied the experience Wantedness value : activates action modules 20

The kernel architecture and visual awareness • KA fits into a model of visual awareness 21

Unstable vision • Well-known ‘Necker cube’ – A ‘reversible’ figure • Feature A is in front of B (or situation reverse) • Once reversal is experienced can neither stop it nor control it 22

Unstable vision • Well-known ‘Necker cube’ – KA • Two major pathways between input and eventual muscular action – Ventral pathway and dorsal one • Ventral path – Carrying signals from visual input that identify what is seen • Dorsal path – A control input to action areas of the brain – Participants with lesioned ventral pathway • Generating appropriate actions in response to visual input • Without reporting any awareness of the visual event • ‘Blind sight’ – Dorsal pathway does not contribute to conscious sensation 23

Unstable vision • The degree of puzzlement of Necker cube illusion 24

Is an organism conscious? • Assertion 3 – There is a prima facie case for an organism to be conscious if it has a system that is isomorphic with the Kernel Architecture hence support the five axiomatic mechanisms • Necessary, not sufficient 25

Animal consciousness • Finding KA in brain – Concomitant with being conscious as – Ka may clearly be seen in invertebrates such as bees • But not in amoebae 26

Higher order thought • High order thought (Rosenthal, 1993) – A sensation becomes conscious only if it is accompanied by a higher level awareness that the perception is happening – In HOT theory, an organism has conscious thoughts only if they occur at two levels 27

Summary • Studying philosophical theme of consciousness – Using computer modeling – Assumption • Consciousness arises from the machinery of the brain and that computational work • Phenomenological approach – Explicit representation of personal sensation – Decomposition of consciousness • Five axioms – Presence in the world – Imagined existence – Choice of experience through attention – Volition and emotion 28

Summary • Depiction – A mix of processing sensory information with muscular information that encodes source information about the sensory information 29